AI Infrastructure
What you must understand
AI infrastructure in Boomi involves connecting processes to AI services, managing credentials, and handling model endpoints. Know how to call an LLM endpoint through an HTTP connector or a dedicated connector, how to pass prompts and context, and how to parse structured responses. Understand secret management, environment-specific configuration, and timeout and retry behavior for slow model calls. Be aware of token limits, cost controls, and data residency considerations. Also understand how AI steps fit into a process: they are just shapes with inputs and outputs, subject to the same error handling as any connector.
Traps that appear
Expect questions on credential scope and whether secrets belong in process properties or a secure store. Another trap is assuming model calls are fast; timeouts and retries matter. Watch for questions about prompt injection and whether user input should be sanitized before reaching a model. Also expect traps around response parsing when a model returns free text instead of JSON.
Worked example
You must summarize incoming tickets. The rule of thumb: validate input, call the model, validate output. Sanitize ticket text, call the endpoint with a bounded prompt, request JSON output, and parse with a fallback if parsing fails. Wrap the call in try/catch with a retry and a timeout, and log the request ID for traceability.
Rule of thumb
Treat every model call as an unreliable external dependency: bound it, retry it, and validate its output.